1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
12 #include <linux/bio.h>
13 #include <linux/bitops.h>
14 #include <linux/blkdev.h>
15 #include <linux/completion.h>
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
23 #include <linux/blk-mq.h>
24 #include <linux/ratelimit.h>
25 #include <asm/unaligned.h>
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_cmnd.h>
29 #include <scsi/scsi_dbg.h>
30 #include <scsi/scsi_device.h>
31 #include <scsi/scsi_driver.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_host.h>
34 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
35 #include <scsi/scsi_dh.h>
37 #include <trace/events/scsi.h>
39 #include "scsi_debugfs.h"
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
44 * Size of integrity metadata is usually small, 1 inline sg should
47 #ifdef CONFIG_ARCH_NO_SG_CHAIN
48 #define SCSI_INLINE_PROT_SG_CNT 0
49 #define SCSI_INLINE_SG_CNT 0
51 #define SCSI_INLINE_PROT_SG_CNT 1
52 #define SCSI_INLINE_SG_CNT 2
55 static struct kmem_cache *scsi_sense_cache;
56 static struct kmem_cache *scsi_sense_isadma_cache;
57 static DEFINE_MUTEX(scsi_sense_cache_mutex);
59 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61 static inline struct kmem_cache *
62 scsi_select_sense_cache(bool unchecked_isa_dma)
64 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
67 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
68 unsigned char *sense_buffer)
70 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
74 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
75 gfp_t gfp_mask, int numa_node)
77 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
81 int scsi_init_sense_cache(struct Scsi_Host *shost)
83 struct kmem_cache *cache;
86 mutex_lock(&scsi_sense_cache_mutex);
87 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
91 if (shost->unchecked_isa_dma) {
92 scsi_sense_isadma_cache =
93 kmem_cache_create("scsi_sense_cache(DMA)",
94 SCSI_SENSE_BUFFERSIZE, 0,
95 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
96 if (!scsi_sense_isadma_cache)
100 kmem_cache_create_usercopy("scsi_sense_cache",
101 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
102 0, SCSI_SENSE_BUFFERSIZE, NULL);
103 if (!scsi_sense_cache)
107 mutex_unlock(&scsi_sense_cache_mutex);
112 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
113 * not change behaviour from the previous unplug mechanism, experimentation
114 * may prove this needs changing.
116 #define SCSI_QUEUE_DELAY 3
119 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
121 struct Scsi_Host *host = cmd->device->host;
122 struct scsi_device *device = cmd->device;
123 struct scsi_target *starget = scsi_target(device);
126 * Set the appropriate busy bit for the device/host.
128 * If the host/device isn't busy, assume that something actually
129 * completed, and that we should be able to queue a command now.
131 * Note that the prior mid-layer assumption that any host could
132 * always queue at least one command is now broken. The mid-layer
133 * will implement a user specifiable stall (see
134 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
135 * if a command is requeued with no other commands outstanding
136 * either for the device or for the host.
139 case SCSI_MLQUEUE_HOST_BUSY:
140 atomic_set(&host->host_blocked, host->max_host_blocked);
142 case SCSI_MLQUEUE_DEVICE_BUSY:
143 case SCSI_MLQUEUE_EH_RETRY:
144 atomic_set(&device->device_blocked,
145 device->max_device_blocked);
147 case SCSI_MLQUEUE_TARGET_BUSY:
148 atomic_set(&starget->target_blocked,
149 starget->max_target_blocked);
154 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
156 if (cmd->request->rq_flags & RQF_DONTPREP) {
157 cmd->request->rq_flags &= ~RQF_DONTPREP;
158 scsi_mq_uninit_cmd(cmd);
162 blk_mq_requeue_request(cmd->request, true);
166 * __scsi_queue_insert - private queue insertion
167 * @cmd: The SCSI command being requeued
168 * @reason: The reason for the requeue
169 * @unbusy: Whether the queue should be unbusied
171 * This is a private queue insertion. The public interface
172 * scsi_queue_insert() always assumes the queue should be unbusied
173 * because it's always called before the completion. This function is
174 * for a requeue after completion, which should only occur in this
177 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
179 struct scsi_device *device = cmd->device;
181 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
182 "Inserting command %p into mlqueue\n", cmd));
184 scsi_set_blocked(cmd, reason);
187 * Decrement the counters, since these commands are no longer
188 * active on the host/device.
191 scsi_device_unbusy(device, cmd);
194 * Requeue this command. It will go before all other commands
195 * that are already in the queue. Schedule requeue work under
196 * lock such that the kblockd_schedule_work() call happens
197 * before blk_cleanup_queue() finishes.
201 blk_mq_requeue_request(cmd->request, true);
205 * scsi_queue_insert - Reinsert a command in the queue.
206 * @cmd: command that we are adding to queue.
207 * @reason: why we are inserting command to queue.
209 * We do this for one of two cases. Either the host is busy and it cannot accept
210 * any more commands for the time being, or the device returned QUEUE_FULL and
211 * can accept no more commands.
213 * Context: This could be called either from an interrupt context or a normal
216 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
218 __scsi_queue_insert(cmd, reason, true);
223 * __scsi_execute - insert request and wait for the result
226 * @data_direction: data direction
227 * @buffer: data buffer
228 * @bufflen: len of buffer
229 * @sense: optional sense buffer
230 * @sshdr: optional decoded sense header
231 * @timeout: request timeout in seconds
232 * @retries: number of times to retry request
233 * @flags: flags for ->cmd_flags
234 * @rq_flags: flags for ->rq_flags
235 * @resid: optional residual length
237 * Returns the scsi_cmnd result field if a command was executed, or a negative
238 * Linux error code if we didn't get that far.
240 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
241 int data_direction, void *buffer, unsigned bufflen,
242 unsigned char *sense, struct scsi_sense_hdr *sshdr,
243 int timeout, int retries, u64 flags, req_flags_t rq_flags,
247 struct scsi_request *rq;
248 int ret = DRIVER_ERROR << 24;
250 req = blk_get_request(sdev->request_queue,
251 data_direction == DMA_TO_DEVICE ?
252 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
257 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
258 buffer, bufflen, GFP_NOIO))
261 rq->cmd_len = COMMAND_SIZE(cmd[0]);
262 memcpy(rq->cmd, cmd, rq->cmd_len);
263 rq->retries = retries;
264 req->timeout = timeout;
265 req->cmd_flags |= flags;
266 req->rq_flags |= rq_flags | RQF_QUIET;
269 * head injection *required* here otherwise quiesce won't work
271 blk_execute_rq(req->q, NULL, req, 1);
274 * Some devices (USB mass-storage in particular) may transfer
275 * garbage data together with a residue indicating that the data
276 * is invalid. Prevent the garbage from being misinterpreted
277 * and prevent security leaks by zeroing out the excess data.
279 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
280 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
283 *resid = rq->resid_len;
284 if (sense && rq->sense_len)
285 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
287 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
290 blk_put_request(req);
294 EXPORT_SYMBOL(__scsi_execute);
297 * Wake up the error handler if necessary. Avoid as follows that the error
298 * handler is not woken up if host in-flight requests number ==
299 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
300 * with an RCU read lock in this function to ensure that this function in
301 * its entirety either finishes before scsi_eh_scmd_add() increases the
302 * host_failed counter or that it notices the shost state change made by
303 * scsi_eh_scmd_add().
305 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
310 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
311 if (unlikely(scsi_host_in_recovery(shost))) {
312 spin_lock_irqsave(shost->host_lock, flags);
313 if (shost->host_failed || shost->host_eh_scheduled)
314 scsi_eh_wakeup(shost);
315 spin_unlock_irqrestore(shost->host_lock, flags);
320 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
322 struct Scsi_Host *shost = sdev->host;
323 struct scsi_target *starget = scsi_target(sdev);
325 scsi_dec_host_busy(shost, cmd);
327 if (starget->can_queue > 0)
328 atomic_dec(&starget->target_busy);
330 atomic_dec(&sdev->device_busy);
333 static void scsi_kick_queue(struct request_queue *q)
335 blk_mq_run_hw_queues(q, false);
339 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
340 * and call blk_run_queue for all the scsi_devices on the target -
341 * including current_sdev first.
343 * Called with *no* scsi locks held.
345 static void scsi_single_lun_run(struct scsi_device *current_sdev)
347 struct Scsi_Host *shost = current_sdev->host;
348 struct scsi_device *sdev, *tmp;
349 struct scsi_target *starget = scsi_target(current_sdev);
352 spin_lock_irqsave(shost->host_lock, flags);
353 starget->starget_sdev_user = NULL;
354 spin_unlock_irqrestore(shost->host_lock, flags);
357 * Call blk_run_queue for all LUNs on the target, starting with
358 * current_sdev. We race with others (to set starget_sdev_user),
359 * but in most cases, we will be first. Ideally, each LU on the
360 * target would get some limited time or requests on the target.
362 scsi_kick_queue(current_sdev->request_queue);
364 spin_lock_irqsave(shost->host_lock, flags);
365 if (starget->starget_sdev_user)
367 list_for_each_entry_safe(sdev, tmp, &starget->devices,
368 same_target_siblings) {
369 if (sdev == current_sdev)
371 if (scsi_device_get(sdev))
374 spin_unlock_irqrestore(shost->host_lock, flags);
375 scsi_kick_queue(sdev->request_queue);
376 spin_lock_irqsave(shost->host_lock, flags);
378 scsi_device_put(sdev);
381 spin_unlock_irqrestore(shost->host_lock, flags);
384 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
386 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
388 if (atomic_read(&sdev->device_blocked) > 0)
393 static inline bool scsi_target_is_busy(struct scsi_target *starget)
395 if (starget->can_queue > 0) {
396 if (atomic_read(&starget->target_busy) >= starget->can_queue)
398 if (atomic_read(&starget->target_blocked) > 0)
404 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
406 if (atomic_read(&shost->host_blocked) > 0)
408 if (shost->host_self_blocked)
413 static void scsi_starved_list_run(struct Scsi_Host *shost)
415 LIST_HEAD(starved_list);
416 struct scsi_device *sdev;
419 spin_lock_irqsave(shost->host_lock, flags);
420 list_splice_init(&shost->starved_list, &starved_list);
422 while (!list_empty(&starved_list)) {
423 struct request_queue *slq;
426 * As long as shost is accepting commands and we have
427 * starved queues, call blk_run_queue. scsi_request_fn
428 * drops the queue_lock and can add us back to the
431 * host_lock protects the starved_list and starved_entry.
432 * scsi_request_fn must get the host_lock before checking
433 * or modifying starved_list or starved_entry.
435 if (scsi_host_is_busy(shost))
438 sdev = list_entry(starved_list.next,
439 struct scsi_device, starved_entry);
440 list_del_init(&sdev->starved_entry);
441 if (scsi_target_is_busy(scsi_target(sdev))) {
442 list_move_tail(&sdev->starved_entry,
443 &shost->starved_list);
448 * Once we drop the host lock, a racing scsi_remove_device()
449 * call may remove the sdev from the starved list and destroy
450 * it and the queue. Mitigate by taking a reference to the
451 * queue and never touching the sdev again after we drop the
452 * host lock. Note: if __scsi_remove_device() invokes
453 * blk_cleanup_queue() before the queue is run from this
454 * function then blk_run_queue() will return immediately since
455 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
457 slq = sdev->request_queue;
458 if (!blk_get_queue(slq))
460 spin_unlock_irqrestore(shost->host_lock, flags);
462 scsi_kick_queue(slq);
465 spin_lock_irqsave(shost->host_lock, flags);
467 /* put any unprocessed entries back */
468 list_splice(&starved_list, &shost->starved_list);
469 spin_unlock_irqrestore(shost->host_lock, flags);
473 * scsi_run_queue - Select a proper request queue to serve next.
474 * @q: last request's queue
476 * The previous command was completely finished, start a new one if possible.
478 static void scsi_run_queue(struct request_queue *q)
480 struct scsi_device *sdev = q->queuedata;
482 if (scsi_target(sdev)->single_lun)
483 scsi_single_lun_run(sdev);
484 if (!list_empty(&sdev->host->starved_list))
485 scsi_starved_list_run(sdev->host);
487 blk_mq_run_hw_queues(q, false);
490 void scsi_requeue_run_queue(struct work_struct *work)
492 struct scsi_device *sdev;
493 struct request_queue *q;
495 sdev = container_of(work, struct scsi_device, requeue_work);
496 q = sdev->request_queue;
500 void scsi_run_host_queues(struct Scsi_Host *shost)
502 struct scsi_device *sdev;
504 shost_for_each_device(sdev, shost)
505 scsi_run_queue(sdev->request_queue);
508 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
510 if (!blk_rq_is_passthrough(cmd->request)) {
511 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
513 if (drv->uninit_command)
514 drv->uninit_command(cmd);
518 void scsi_free_sgtables(struct scsi_cmnd *cmd)
520 if (cmd->sdb.table.nents)
521 sg_free_table_chained(&cmd->sdb.table,
523 if (scsi_prot_sg_count(cmd))
524 sg_free_table_chained(&cmd->prot_sdb->table,
525 SCSI_INLINE_PROT_SG_CNT);
527 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
529 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
531 scsi_free_sgtables(cmd);
532 scsi_uninit_cmd(cmd);
535 static void scsi_run_queue_async(struct scsi_device *sdev)
537 if (scsi_target(sdev)->single_lun ||
538 !list_empty(&sdev->host->starved_list)) {
539 kblockd_schedule_work(&sdev->requeue_work);
542 * smp_mb() present in sbitmap_queue_clear() or implied in
543 * .end_io is for ordering writing .device_busy in
544 * scsi_device_unbusy() and reading sdev->restarts.
546 int old = atomic_read(&sdev->restarts);
549 * ->restarts has to be kept as non-zero if new budget
552 * No need to run queue when either another re-run
553 * queue wins in updating ->restarts or a new budget
556 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
557 blk_mq_run_hw_queues(sdev->request_queue, true);
561 /* Returns false when no more bytes to process, true if there are more */
562 static bool scsi_end_request(struct request *req, blk_status_t error,
565 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
566 struct scsi_device *sdev = cmd->device;
567 struct request_queue *q = sdev->request_queue;
569 if (blk_update_request(req, error, bytes))
572 if (blk_queue_add_random(q))
573 add_disk_randomness(req->rq_disk);
575 if (!blk_rq_is_scsi(req)) {
576 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
577 cmd->flags &= ~SCMD_INITIALIZED;
581 * Calling rcu_barrier() is not necessary here because the
582 * SCSI error handler guarantees that the function called by
583 * call_rcu() has been called before scsi_end_request() is
586 destroy_rcu_head(&cmd->rcu);
589 * In the MQ case the command gets freed by __blk_mq_end_request,
590 * so we have to do all cleanup that depends on it earlier.
592 * We also can't kick the queues from irq context, so we
593 * will have to defer it to a workqueue.
595 scsi_mq_uninit_cmd(cmd);
598 * queue is still alive, so grab the ref for preventing it
599 * from being cleaned up during running queue.
601 percpu_ref_get(&q->q_usage_counter);
603 __blk_mq_end_request(req, error);
605 scsi_run_queue_async(sdev);
607 percpu_ref_put(&q->q_usage_counter);
612 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
614 * @result: scsi error code
616 * Translate a SCSI result code into a blk_status_t value. May reset the host
617 * byte of @cmd->result.
619 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
621 switch (host_byte(result)) {
624 * Also check the other bytes than the status byte in result
625 * to handle the case when a SCSI LLD sets result to
626 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
628 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
630 return BLK_STS_IOERR;
631 case DID_TRANSPORT_FAILFAST:
632 return BLK_STS_TRANSPORT;
633 case DID_TARGET_FAILURE:
634 set_host_byte(cmd, DID_OK);
635 return BLK_STS_TARGET;
636 case DID_NEXUS_FAILURE:
637 set_host_byte(cmd, DID_OK);
638 return BLK_STS_NEXUS;
639 case DID_ALLOC_FAILURE:
640 set_host_byte(cmd, DID_OK);
641 return BLK_STS_NOSPC;
642 case DID_MEDIUM_ERROR:
643 set_host_byte(cmd, DID_OK);
644 return BLK_STS_MEDIUM;
646 return BLK_STS_IOERR;
650 /* Helper for scsi_io_completion() when "reprep" action required. */
651 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
652 struct request_queue *q)
654 /* A new command will be prepared and issued. */
655 scsi_mq_requeue_cmd(cmd);
658 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
660 struct request *req = cmd->request;
661 unsigned long wait_for;
663 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
666 wait_for = (cmd->allowed + 1) * req->timeout;
667 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
668 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
675 /* Helper for scsi_io_completion() when special action required. */
676 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
678 struct request_queue *q = cmd->device->request_queue;
679 struct request *req = cmd->request;
681 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
682 ACTION_DELAYED_RETRY} action;
683 struct scsi_sense_hdr sshdr;
685 bool sense_current = true; /* false implies "deferred sense" */
686 blk_status_t blk_stat;
688 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
690 sense_current = !scsi_sense_is_deferred(&sshdr);
692 blk_stat = scsi_result_to_blk_status(cmd, result);
694 if (host_byte(result) == DID_RESET) {
695 /* Third party bus reset or reset for error recovery
696 * reasons. Just retry the command and see what
699 action = ACTION_RETRY;
700 } else if (sense_valid && sense_current) {
701 switch (sshdr.sense_key) {
703 if (cmd->device->removable) {
704 /* Detected disc change. Set a bit
705 * and quietly refuse further access.
707 cmd->device->changed = 1;
708 action = ACTION_FAIL;
710 /* Must have been a power glitch, or a
711 * bus reset. Could not have been a
712 * media change, so we just retry the
713 * command and see what happens.
715 action = ACTION_RETRY;
718 case ILLEGAL_REQUEST:
719 /* If we had an ILLEGAL REQUEST returned, then
720 * we may have performed an unsupported
721 * command. The only thing this should be
722 * would be a ten byte read where only a six
723 * byte read was supported. Also, on a system
724 * where READ CAPACITY failed, we may have
725 * read past the end of the disk.
727 if ((cmd->device->use_10_for_rw &&
728 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
729 (cmd->cmnd[0] == READ_10 ||
730 cmd->cmnd[0] == WRITE_10)) {
731 /* This will issue a new 6-byte command. */
732 cmd->device->use_10_for_rw = 0;
733 action = ACTION_REPREP;
734 } else if (sshdr.asc == 0x10) /* DIX */ {
735 action = ACTION_FAIL;
736 blk_stat = BLK_STS_PROTECTION;
737 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
738 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
739 action = ACTION_FAIL;
740 blk_stat = BLK_STS_TARGET;
742 action = ACTION_FAIL;
744 case ABORTED_COMMAND:
745 action = ACTION_FAIL;
746 if (sshdr.asc == 0x10) /* DIF */
747 blk_stat = BLK_STS_PROTECTION;
750 /* If the device is in the process of becoming
751 * ready, or has a temporary blockage, retry.
753 if (sshdr.asc == 0x04) {
754 switch (sshdr.ascq) {
755 case 0x01: /* becoming ready */
756 case 0x04: /* format in progress */
757 case 0x05: /* rebuild in progress */
758 case 0x06: /* recalculation in progress */
759 case 0x07: /* operation in progress */
760 case 0x08: /* Long write in progress */
761 case 0x09: /* self test in progress */
762 case 0x14: /* space allocation in progress */
763 case 0x1a: /* start stop unit in progress */
764 case 0x1b: /* sanitize in progress */
765 case 0x1d: /* configuration in progress */
766 case 0x24: /* depopulation in progress */
767 action = ACTION_DELAYED_RETRY;
770 action = ACTION_FAIL;
774 action = ACTION_FAIL;
776 case VOLUME_OVERFLOW:
777 /* See SSC3rXX or current. */
778 action = ACTION_FAIL;
781 action = ACTION_FAIL;
782 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
783 (sshdr.asc == 0x55 &&
784 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
785 /* Insufficient zone resources */
786 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
790 action = ACTION_FAIL;
794 action = ACTION_FAIL;
796 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
797 action = ACTION_FAIL;
801 /* Give up and fail the remainder of the request */
802 if (!(req->rq_flags & RQF_QUIET)) {
803 static DEFINE_RATELIMIT_STATE(_rs,
804 DEFAULT_RATELIMIT_INTERVAL,
805 DEFAULT_RATELIMIT_BURST);
807 if (unlikely(scsi_logging_level))
809 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
810 SCSI_LOG_MLCOMPLETE_BITS);
813 * if logging is enabled the failure will be printed
814 * in scsi_log_completion(), so avoid duplicate messages
816 if (!level && __ratelimit(&_rs)) {
817 scsi_print_result(cmd, NULL, FAILED);
818 if (driver_byte(result) == DRIVER_SENSE)
819 scsi_print_sense(cmd);
820 scsi_print_command(cmd);
823 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
827 scsi_io_completion_reprep(cmd, q);
830 /* Retry the same command immediately */
831 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
833 case ACTION_DELAYED_RETRY:
834 /* Retry the same command after a delay */
835 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
841 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
842 * new result that may suppress further error checking. Also modifies
843 * *blk_statp in some cases.
845 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
846 blk_status_t *blk_statp)
849 bool sense_current = true; /* false implies "deferred sense" */
850 struct request *req = cmd->request;
851 struct scsi_sense_hdr sshdr;
853 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
855 sense_current = !scsi_sense_is_deferred(&sshdr);
857 if (blk_rq_is_passthrough(req)) {
860 * SG_IO wants current and deferred errors
862 scsi_req(req)->sense_len =
863 min(8 + cmd->sense_buffer[7],
864 SCSI_SENSE_BUFFERSIZE);
867 *blk_statp = scsi_result_to_blk_status(cmd, result);
868 } else if (blk_rq_bytes(req) == 0 && sense_current) {
870 * Flush commands do not transfers any data, and thus cannot use
871 * good_bytes != blk_rq_bytes(req) as the signal for an error.
872 * This sets *blk_statp explicitly for the problem case.
874 *blk_statp = scsi_result_to_blk_status(cmd, result);
877 * Recovered errors need reporting, but they're always treated as
878 * success, so fiddle the result code here. For passthrough requests
879 * we already took a copy of the original into sreq->result which
880 * is what gets returned to the user
882 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
883 bool do_print = true;
885 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
886 * skip print since caller wants ATA registers. Only occurs
887 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
889 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
891 else if (req->rq_flags & RQF_QUIET)
894 scsi_print_sense(cmd);
896 /* for passthrough, *blk_statp may be set */
897 *blk_statp = BLK_STS_OK;
900 * Another corner case: the SCSI status byte is non-zero but 'good'.
901 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
902 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
903 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
904 * intermediate statuses (both obsolete in SAM-4) as good.
906 if (status_byte(result) && scsi_status_is_good(result)) {
908 *blk_statp = BLK_STS_OK;
914 * scsi_io_completion - Completion processing for SCSI commands.
915 * @cmd: command that is finished.
916 * @good_bytes: number of processed bytes.
918 * We will finish off the specified number of sectors. If we are done, the
919 * command block will be released and the queue function will be goosed. If we
920 * are not done then we have to figure out what to do next:
922 * a) We can call scsi_io_completion_reprep(). The request will be
923 * unprepared and put back on the queue. Then a new command will
924 * be created for it. This should be used if we made forward
925 * progress, or if we want to switch from READ(10) to READ(6) for
928 * b) We can call scsi_io_completion_action(). The request will be
929 * put back on the queue and retried using the same command as
930 * before, possibly after a delay.
932 * c) We can call scsi_end_request() with blk_stat other than
933 * BLK_STS_OK, to fail the remainder of the request.
935 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
937 int result = cmd->result;
938 struct request_queue *q = cmd->device->request_queue;
939 struct request *req = cmd->request;
940 blk_status_t blk_stat = BLK_STS_OK;
942 if (unlikely(result)) /* a nz result may or may not be an error */
943 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
945 if (unlikely(blk_rq_is_passthrough(req))) {
947 * scsi_result_to_blk_status may have reset the host_byte
949 scsi_req(req)->result = cmd->result;
953 * Next deal with any sectors which we were able to correctly
956 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
957 "%u sectors total, %d bytes done.\n",
958 blk_rq_sectors(req), good_bytes));
961 * Failed, zero length commands always need to drop down
962 * to retry code. Fast path should return in this block.
964 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
965 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
966 return; /* no bytes remaining */
969 /* Kill remainder if no retries. */
970 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
971 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
973 "Bytes remaining after failed, no-retry command");
978 * If there had been no error, but we have leftover bytes in the
979 * requeues just queue the command up again.
981 if (likely(result == 0))
982 scsi_io_completion_reprep(cmd, q);
984 scsi_io_completion_action(cmd, result);
987 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
990 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
991 !op_is_write(req_op(rq)) &&
992 sdev->host->hostt->dma_need_drain(rq);
996 * scsi_alloc_sgtables - allocate S/G tables for a command
997 * @cmd: command descriptor we wish to initialize
1000 * * BLK_STS_OK - on success
1001 * * BLK_STS_RESOURCE - if the failure is retryable
1002 * * BLK_STS_IOERR - if the failure is fatal
1004 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1006 struct scsi_device *sdev = cmd->device;
1007 struct request *rq = cmd->request;
1008 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1009 struct scatterlist *last_sg = NULL;
1011 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1014 if (WARN_ON_ONCE(!nr_segs))
1015 return BLK_STS_IOERR;
1018 * Make sure there is space for the drain. The driver must adjust
1019 * max_hw_segments to be prepared for this.
1025 * If sg table allocation fails, requeue request later.
1027 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1028 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1029 return BLK_STS_RESOURCE;
1032 * Next, walk the list, and fill in the addresses and sizes of
1035 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1037 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1038 unsigned int pad_len =
1039 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1041 last_sg->length += pad_len;
1042 cmd->extra_len += pad_len;
1046 sg_unmark_end(last_sg);
1047 last_sg = sg_next(last_sg);
1048 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1049 sg_mark_end(last_sg);
1051 cmd->extra_len += sdev->dma_drain_len;
1055 BUG_ON(count > cmd->sdb.table.nents);
1056 cmd->sdb.table.nents = count;
1057 cmd->sdb.length = blk_rq_payload_bytes(rq);
1059 if (blk_integrity_rq(rq)) {
1060 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1063 if (WARN_ON_ONCE(!prot_sdb)) {
1065 * This can happen if someone (e.g. multipath)
1066 * queues a command to a device on an adapter
1067 * that does not support DIX.
1069 ret = BLK_STS_IOERR;
1070 goto out_free_sgtables;
1073 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1075 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1076 prot_sdb->table.sgl,
1077 SCSI_INLINE_PROT_SG_CNT)) {
1078 ret = BLK_STS_RESOURCE;
1079 goto out_free_sgtables;
1082 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1083 prot_sdb->table.sgl);
1084 BUG_ON(count > ivecs);
1085 BUG_ON(count > queue_max_integrity_segments(rq->q));
1087 cmd->prot_sdb = prot_sdb;
1088 cmd->prot_sdb->table.nents = count;
1093 scsi_free_sgtables(cmd);
1096 EXPORT_SYMBOL(scsi_alloc_sgtables);
1099 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1100 * @rq: Request associated with the SCSI command to be initialized.
1102 * This function initializes the members of struct scsi_cmnd that must be
1103 * initialized before request processing starts and that won't be
1104 * reinitialized if a SCSI command is requeued.
1106 * Called from inside blk_get_request() for pass-through requests and from
1107 * inside scsi_init_command() for filesystem requests.
1109 static void scsi_initialize_rq(struct request *rq)
1111 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1113 scsi_req_init(&cmd->req);
1114 init_rcu_head(&cmd->rcu);
1115 cmd->jiffies_at_alloc = jiffies;
1120 * Only called when the request isn't completed by SCSI, and not freed by
1123 static void scsi_cleanup_rq(struct request *rq)
1125 if (rq->rq_flags & RQF_DONTPREP) {
1126 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1127 rq->rq_flags &= ~RQF_DONTPREP;
1131 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1132 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1134 void *buf = cmd->sense_buffer;
1135 void *prot = cmd->prot_sdb;
1136 struct request *rq = blk_mq_rq_from_pdu(cmd);
1137 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1138 unsigned long jiffies_at_alloc;
1139 int retries, to_clear;
1142 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1143 flags |= SCMD_INITIALIZED;
1144 scsi_initialize_rq(rq);
1147 jiffies_at_alloc = cmd->jiffies_at_alloc;
1148 retries = cmd->retries;
1149 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1151 * Zero out the cmd, except for the embedded scsi_request. Only clear
1152 * the driver-private command data if the LLD does not supply a
1153 * function to initialize that data.
1155 to_clear = sizeof(*cmd) - sizeof(cmd->req);
1156 if (!dev->host->hostt->init_cmd_priv)
1157 to_clear += dev->host->hostt->cmd_size;
1158 memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1161 cmd->sense_buffer = buf;
1162 cmd->prot_sdb = prot;
1164 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1165 cmd->jiffies_at_alloc = jiffies_at_alloc;
1166 cmd->retries = retries;
1168 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1172 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1173 struct request *req)
1175 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1178 * Passthrough requests may transfer data, in which case they must
1179 * a bio attached to them. Or they might contain a SCSI command
1180 * that does not transfer data, in which case they may optionally
1181 * submit a request without an attached bio.
1184 blk_status_t ret = scsi_alloc_sgtables(cmd);
1185 if (unlikely(ret != BLK_STS_OK))
1188 BUG_ON(blk_rq_bytes(req));
1190 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1193 cmd->cmd_len = scsi_req(req)->cmd_len;
1194 if (cmd->cmd_len == 0)
1195 cmd->cmd_len = scsi_command_size(cmd->cmnd);
1196 cmd->cmnd = scsi_req(req)->cmd;
1197 cmd->transfersize = blk_rq_bytes(req);
1198 cmd->allowed = scsi_req(req)->retries;
1203 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1205 switch (sdev->sdev_state) {
1207 case SDEV_TRANSPORT_OFFLINE:
1209 * If the device is offline we refuse to process any
1210 * commands. The device must be brought online
1211 * before trying any recovery commands.
1213 if (!sdev->offline_already) {
1214 sdev->offline_already = true;
1215 sdev_printk(KERN_ERR, sdev,
1216 "rejecting I/O to offline device\n");
1218 return BLK_STS_IOERR;
1221 * If the device is fully deleted, we refuse to
1222 * process any commands as well.
1224 sdev_printk(KERN_ERR, sdev,
1225 "rejecting I/O to dead device\n");
1226 return BLK_STS_IOERR;
1228 case SDEV_CREATED_BLOCK:
1229 return BLK_STS_RESOURCE;
1232 * If the devices is blocked we defer normal commands.
1234 if (req && !(req->rq_flags & RQF_PREEMPT))
1235 return BLK_STS_RESOURCE;
1239 * For any other not fully online state we only allow
1240 * special commands. In particular any user initiated
1241 * command is not allowed.
1243 if (req && !(req->rq_flags & RQF_PREEMPT))
1244 return BLK_STS_IOERR;
1250 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1253 * Called with the queue_lock held.
1255 static inline int scsi_dev_queue_ready(struct request_queue *q,
1256 struct scsi_device *sdev)
1260 busy = atomic_inc_return(&sdev->device_busy) - 1;
1261 if (atomic_read(&sdev->device_blocked)) {
1266 * unblock after device_blocked iterates to zero
1268 if (atomic_dec_return(&sdev->device_blocked) > 0)
1270 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1271 "unblocking device at zero depth\n"));
1274 if (busy >= sdev->queue_depth)
1279 atomic_dec(&sdev->device_busy);
1284 * scsi_target_queue_ready: checks if there we can send commands to target
1285 * @sdev: scsi device on starget to check.
1287 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1288 struct scsi_device *sdev)
1290 struct scsi_target *starget = scsi_target(sdev);
1293 if (starget->single_lun) {
1294 spin_lock_irq(shost->host_lock);
1295 if (starget->starget_sdev_user &&
1296 starget->starget_sdev_user != sdev) {
1297 spin_unlock_irq(shost->host_lock);
1300 starget->starget_sdev_user = sdev;
1301 spin_unlock_irq(shost->host_lock);
1304 if (starget->can_queue <= 0)
1307 busy = atomic_inc_return(&starget->target_busy) - 1;
1308 if (atomic_read(&starget->target_blocked) > 0) {
1313 * unblock after target_blocked iterates to zero
1315 if (atomic_dec_return(&starget->target_blocked) > 0)
1318 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1319 "unblocking target at zero depth\n"));
1322 if (busy >= starget->can_queue)
1328 spin_lock_irq(shost->host_lock);
1329 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1330 spin_unlock_irq(shost->host_lock);
1332 if (starget->can_queue > 0)
1333 atomic_dec(&starget->target_busy);
1338 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1339 * return 0. We must end up running the queue again whenever 0 is
1340 * returned, else IO can hang.
1342 static inline int scsi_host_queue_ready(struct request_queue *q,
1343 struct Scsi_Host *shost,
1344 struct scsi_device *sdev,
1345 struct scsi_cmnd *cmd)
1347 if (scsi_host_in_recovery(shost))
1350 if (atomic_read(&shost->host_blocked) > 0) {
1351 if (scsi_host_busy(shost) > 0)
1355 * unblock after host_blocked iterates to zero
1357 if (atomic_dec_return(&shost->host_blocked) > 0)
1361 shost_printk(KERN_INFO, shost,
1362 "unblocking host at zero depth\n"));
1365 if (shost->host_self_blocked)
1368 /* We're OK to process the command, so we can't be starved */
1369 if (!list_empty(&sdev->starved_entry)) {
1370 spin_lock_irq(shost->host_lock);
1371 if (!list_empty(&sdev->starved_entry))
1372 list_del_init(&sdev->starved_entry);
1373 spin_unlock_irq(shost->host_lock);
1376 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1381 spin_lock_irq(shost->host_lock);
1382 if (list_empty(&sdev->starved_entry))
1383 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1384 spin_unlock_irq(shost->host_lock);
1386 scsi_dec_host_busy(shost, cmd);
1391 * Busy state exporting function for request stacking drivers.
1393 * For efficiency, no lock is taken to check the busy state of
1394 * shost/starget/sdev, since the returned value is not guaranteed and
1395 * may be changed after request stacking drivers call the function,
1396 * regardless of taking lock or not.
1398 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1399 * needs to return 'not busy'. Otherwise, request stacking drivers
1400 * may hold requests forever.
1402 static bool scsi_mq_lld_busy(struct request_queue *q)
1404 struct scsi_device *sdev = q->queuedata;
1405 struct Scsi_Host *shost;
1407 if (blk_queue_dying(q))
1413 * Ignore host/starget busy state.
1414 * Since block layer does not have a concept of fairness across
1415 * multiple queues, congestion of host/starget needs to be handled
1418 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1424 static void scsi_softirq_done(struct request *rq)
1426 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1429 INIT_LIST_HEAD(&cmd->eh_entry);
1431 atomic_inc(&cmd->device->iodone_cnt);
1433 atomic_inc(&cmd->device->ioerr_cnt);
1435 disposition = scsi_decide_disposition(cmd);
1436 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1437 disposition = SUCCESS;
1439 scsi_log_completion(cmd, disposition);
1441 switch (disposition) {
1443 scsi_finish_command(cmd);
1446 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1448 case ADD_TO_MLQUEUE:
1449 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1452 scsi_eh_scmd_add(cmd);
1458 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1459 * @cmd: command block we are dispatching.
1461 * Return: nonzero return request was rejected and device's queue needs to be
1464 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1466 struct Scsi_Host *host = cmd->device->host;
1469 atomic_inc(&cmd->device->iorequest_cnt);
1471 /* check if the device is still usable */
1472 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1473 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1474 * returns an immediate error upwards, and signals
1475 * that the device is no longer present */
1476 cmd->result = DID_NO_CONNECT << 16;
1480 /* Check to see if the scsi lld made this device blocked. */
1481 if (unlikely(scsi_device_blocked(cmd->device))) {
1483 * in blocked state, the command is just put back on
1484 * the device queue. The suspend state has already
1485 * blocked the queue so future requests should not
1486 * occur until the device transitions out of the
1489 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1490 "queuecommand : device blocked\n"));
1491 return SCSI_MLQUEUE_DEVICE_BUSY;
1494 /* Store the LUN value in cmnd, if needed. */
1495 if (cmd->device->lun_in_cdb)
1496 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1497 (cmd->device->lun << 5 & 0xe0);
1502 * Before we queue this command, check if the command
1503 * length exceeds what the host adapter can handle.
1505 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1506 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1507 "queuecommand : command too long. "
1508 "cdb_size=%d host->max_cmd_len=%d\n",
1509 cmd->cmd_len, cmd->device->host->max_cmd_len));
1510 cmd->result = (DID_ABORT << 16);
1514 if (unlikely(host->shost_state == SHOST_DEL)) {
1515 cmd->result = (DID_NO_CONNECT << 16);
1520 trace_scsi_dispatch_cmd_start(cmd);
1521 rtn = host->hostt->queuecommand(host, cmd);
1523 trace_scsi_dispatch_cmd_error(cmd, rtn);
1524 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1525 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1526 rtn = SCSI_MLQUEUE_HOST_BUSY;
1528 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1529 "queuecommand : request rejected\n"));
1534 cmd->scsi_done(cmd);
1538 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1539 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1541 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1542 sizeof(struct scatterlist);
1545 static blk_status_t scsi_prepare_cmd(struct request *req)
1547 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1548 struct scsi_device *sdev = req->q->queuedata;
1549 struct Scsi_Host *shost = sdev->host;
1550 struct scatterlist *sg;
1552 scsi_init_command(sdev, cmd);
1555 cmd->tag = req->tag;
1556 cmd->prot_op = SCSI_PROT_NORMAL;
1557 if (blk_rq_bytes(req))
1558 cmd->sc_data_direction = rq_dma_dir(req);
1560 cmd->sc_data_direction = DMA_NONE;
1562 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1563 cmd->sdb.table.sgl = sg;
1565 if (scsi_host_get_prot(shost)) {
1566 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1568 cmd->prot_sdb->table.sgl =
1569 (struct scatterlist *)(cmd->prot_sdb + 1);
1573 * Special handling for passthrough commands, which don't go to the ULP
1576 if (blk_rq_is_scsi(req))
1577 return scsi_setup_scsi_cmnd(sdev, req);
1579 if (sdev->handler && sdev->handler->prep_fn) {
1580 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1582 if (ret != BLK_STS_OK)
1586 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1587 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1588 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1591 static void scsi_mq_done(struct scsi_cmnd *cmd)
1593 if (unlikely(blk_should_fake_timeout(cmd->request->q)))
1595 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1597 trace_scsi_dispatch_cmd_done(cmd);
1598 blk_mq_complete_request(cmd->request);
1601 static void scsi_mq_put_budget(struct request_queue *q)
1603 struct scsi_device *sdev = q->queuedata;
1605 atomic_dec(&sdev->device_busy);
1608 static bool scsi_mq_get_budget(struct request_queue *q)
1610 struct scsi_device *sdev = q->queuedata;
1612 if (scsi_dev_queue_ready(q, sdev))
1615 atomic_inc(&sdev->restarts);
1618 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1619 * .restarts must be incremented before .device_busy is read because the
1620 * code in scsi_run_queue_async() depends on the order of these operations.
1622 smp_mb__after_atomic();
1625 * If all in-flight requests originated from this LUN are completed
1626 * before reading .device_busy, sdev->device_busy will be observed as
1627 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1628 * soon. Otherwise, completion of one of these requests will observe
1629 * the .restarts flag, and the request queue will be run for handling
1630 * this request, see scsi_end_request().
1632 if (unlikely(atomic_read(&sdev->device_busy) == 0 &&
1633 !scsi_device_blocked(sdev)))
1634 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1638 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1639 const struct blk_mq_queue_data *bd)
1641 struct request *req = bd->rq;
1642 struct request_queue *q = req->q;
1643 struct scsi_device *sdev = q->queuedata;
1644 struct Scsi_Host *shost = sdev->host;
1645 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1650 * If the device is not in running state we will reject some or all
1653 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1654 ret = scsi_device_state_check(sdev, req);
1655 if (ret != BLK_STS_OK)
1656 goto out_put_budget;
1659 ret = BLK_STS_RESOURCE;
1660 if (!scsi_target_queue_ready(shost, sdev))
1661 goto out_put_budget;
1662 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1663 goto out_dec_target_busy;
1665 if (!(req->rq_flags & RQF_DONTPREP)) {
1666 ret = scsi_prepare_cmd(req);
1667 if (ret != BLK_STS_OK)
1668 goto out_dec_host_busy;
1669 req->rq_flags |= RQF_DONTPREP;
1671 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1674 cmd->flags &= SCMD_PRESERVED_FLAGS;
1675 if (sdev->simple_tags)
1676 cmd->flags |= SCMD_TAGGED;
1678 cmd->flags |= SCMD_LAST;
1680 scsi_set_resid(cmd, 0);
1681 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1682 cmd->scsi_done = scsi_mq_done;
1684 blk_mq_start_request(req);
1685 reason = scsi_dispatch_cmd(cmd);
1687 scsi_set_blocked(cmd, reason);
1688 ret = BLK_STS_RESOURCE;
1689 goto out_dec_host_busy;
1695 scsi_dec_host_busy(shost, cmd);
1696 out_dec_target_busy:
1697 if (scsi_target(sdev)->can_queue > 0)
1698 atomic_dec(&scsi_target(sdev)->target_busy);
1700 scsi_mq_put_budget(q);
1704 case BLK_STS_RESOURCE:
1705 case BLK_STS_ZONE_RESOURCE:
1706 if (atomic_read(&sdev->device_busy) ||
1707 scsi_device_blocked(sdev))
1708 ret = BLK_STS_DEV_RESOURCE;
1711 if (unlikely(!scsi_device_online(sdev)))
1712 scsi_req(req)->result = DID_NO_CONNECT << 16;
1714 scsi_req(req)->result = DID_ERROR << 16;
1716 * Make sure to release all allocated resources when
1717 * we hit an error, as we will never see this command
1720 if (req->rq_flags & RQF_DONTPREP)
1721 scsi_mq_uninit_cmd(cmd);
1722 scsi_run_queue_async(sdev);
1728 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1732 return BLK_EH_RESET_TIMER;
1733 return scsi_times_out(req);
1736 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1737 unsigned int hctx_idx, unsigned int numa_node)
1739 struct Scsi_Host *shost = set->driver_data;
1740 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1741 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1742 struct scatterlist *sg;
1745 if (unchecked_isa_dma)
1746 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1747 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1748 GFP_KERNEL, numa_node);
1749 if (!cmd->sense_buffer)
1751 cmd->req.sense = cmd->sense_buffer;
1753 if (scsi_host_get_prot(shost)) {
1754 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1755 shost->hostt->cmd_size;
1756 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1759 if (shost->hostt->init_cmd_priv) {
1760 ret = shost->hostt->init_cmd_priv(shost, cmd);
1762 scsi_free_sense_buffer(unchecked_isa_dma,
1769 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1770 unsigned int hctx_idx)
1772 struct Scsi_Host *shost = set->driver_data;
1773 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1775 if (shost->hostt->exit_cmd_priv)
1776 shost->hostt->exit_cmd_priv(shost, cmd);
1777 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1781 static int scsi_map_queues(struct blk_mq_tag_set *set)
1783 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1785 if (shost->hostt->map_queues)
1786 return shost->hostt->map_queues(shost);
1787 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1790 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1792 struct device *dev = shost->dma_dev;
1795 * this limit is imposed by hardware restrictions
1797 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1800 if (scsi_host_prot_dma(shost)) {
1801 shost->sg_prot_tablesize =
1802 min_not_zero(shost->sg_prot_tablesize,
1803 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1804 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1805 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1808 if (dev->dma_mask) {
1809 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1810 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1812 blk_queue_max_hw_sectors(q, shost->max_sectors);
1813 if (shost->unchecked_isa_dma)
1814 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1815 blk_queue_segment_boundary(q, shost->dma_boundary);
1816 dma_set_seg_boundary(dev, shost->dma_boundary);
1818 blk_queue_max_segment_size(q, shost->max_segment_size);
1819 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1820 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1823 * Set a reasonable default alignment: The larger of 32-byte (dword),
1824 * which is a common minimum for HBAs, and the minimum DMA alignment,
1825 * which is set by the platform.
1827 * Devices that require a bigger alignment can increase it later.
1829 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1831 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1833 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1834 .get_budget = scsi_mq_get_budget,
1835 .put_budget = scsi_mq_put_budget,
1836 .queue_rq = scsi_queue_rq,
1837 .complete = scsi_softirq_done,
1838 .timeout = scsi_timeout,
1839 #ifdef CONFIG_BLK_DEBUG_FS
1840 .show_rq = scsi_show_rq,
1842 .init_request = scsi_mq_init_request,
1843 .exit_request = scsi_mq_exit_request,
1844 .initialize_rq_fn = scsi_initialize_rq,
1845 .cleanup_rq = scsi_cleanup_rq,
1846 .busy = scsi_mq_lld_busy,
1847 .map_queues = scsi_map_queues,
1851 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1853 struct request_queue *q = hctx->queue;
1854 struct scsi_device *sdev = q->queuedata;
1855 struct Scsi_Host *shost = sdev->host;
1857 shost->hostt->commit_rqs(shost, hctx->queue_num);
1860 static const struct blk_mq_ops scsi_mq_ops = {
1861 .get_budget = scsi_mq_get_budget,
1862 .put_budget = scsi_mq_put_budget,
1863 .queue_rq = scsi_queue_rq,
1864 .commit_rqs = scsi_commit_rqs,
1865 .complete = scsi_softirq_done,
1866 .timeout = scsi_timeout,
1867 #ifdef CONFIG_BLK_DEBUG_FS
1868 .show_rq = scsi_show_rq,
1870 .init_request = scsi_mq_init_request,
1871 .exit_request = scsi_mq_exit_request,
1872 .initialize_rq_fn = scsi_initialize_rq,
1873 .cleanup_rq = scsi_cleanup_rq,
1874 .busy = scsi_mq_lld_busy,
1875 .map_queues = scsi_map_queues,
1878 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1880 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1881 if (IS_ERR(sdev->request_queue))
1884 sdev->request_queue->queuedata = sdev;
1885 __scsi_init_queue(sdev->host, sdev->request_queue);
1886 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1887 return sdev->request_queue;
1890 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1892 unsigned int cmd_size, sgl_size;
1893 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1895 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1896 scsi_mq_inline_sgl_size(shost));
1897 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1898 if (scsi_host_get_prot(shost))
1899 cmd_size += sizeof(struct scsi_data_buffer) +
1900 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1902 memset(tag_set, 0, sizeof(*tag_set));
1903 if (shost->hostt->commit_rqs)
1904 tag_set->ops = &scsi_mq_ops;
1906 tag_set->ops = &scsi_mq_ops_no_commit;
1907 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1908 tag_set->queue_depth = shost->can_queue;
1909 tag_set->cmd_size = cmd_size;
1910 tag_set->numa_node = NUMA_NO_NODE;
1911 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1913 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1914 tag_set->driver_data = shost;
1915 if (shost->host_tagset)
1916 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1918 return blk_mq_alloc_tag_set(tag_set);
1921 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1923 blk_mq_free_tag_set(&shost->tag_set);
1927 * scsi_device_from_queue - return sdev associated with a request_queue
1928 * @q: The request queue to return the sdev from
1930 * Return the sdev associated with a request queue or NULL if the
1931 * request_queue does not reference a SCSI device.
1933 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1935 struct scsi_device *sdev = NULL;
1937 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1938 q->mq_ops == &scsi_mq_ops)
1939 sdev = q->queuedata;
1940 if (!sdev || !get_device(&sdev->sdev_gendev))
1947 * scsi_block_requests - Utility function used by low-level drivers to prevent
1948 * further commands from being queued to the device.
1949 * @shost: host in question
1951 * There is no timer nor any other means by which the requests get unblocked
1952 * other than the low-level driver calling scsi_unblock_requests().
1954 void scsi_block_requests(struct Scsi_Host *shost)
1956 shost->host_self_blocked = 1;
1958 EXPORT_SYMBOL(scsi_block_requests);
1961 * scsi_unblock_requests - Utility function used by low-level drivers to allow
1962 * further commands to be queued to the device.
1963 * @shost: host in question
1965 * There is no timer nor any other means by which the requests get unblocked
1966 * other than the low-level driver calling scsi_unblock_requests(). This is done
1967 * as an API function so that changes to the internals of the scsi mid-layer
1968 * won't require wholesale changes to drivers that use this feature.
1970 void scsi_unblock_requests(struct Scsi_Host *shost)
1972 shost->host_self_blocked = 0;
1973 scsi_run_host_queues(shost);
1975 EXPORT_SYMBOL(scsi_unblock_requests);
1977 void scsi_exit_queue(void)
1979 kmem_cache_destroy(scsi_sense_cache);
1980 kmem_cache_destroy(scsi_sense_isadma_cache);
1984 * scsi_mode_select - issue a mode select
1985 * @sdev: SCSI device to be queried
1986 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1987 * @sp: Save page bit (0 == don't save, 1 == save)
1988 * @modepage: mode page being requested
1989 * @buffer: request buffer (may not be smaller than eight bytes)
1990 * @len: length of request buffer.
1991 * @timeout: command timeout
1992 * @retries: number of retries before failing
1993 * @data: returns a structure abstracting the mode header data
1994 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1995 * must be SCSI_SENSE_BUFFERSIZE big.
1997 * Returns zero if successful; negative error number or scsi
2002 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2003 unsigned char *buffer, int len, int timeout, int retries,
2004 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2006 unsigned char cmd[10];
2007 unsigned char *real_buffer;
2010 memset(cmd, 0, sizeof(cmd));
2011 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2013 if (sdev->use_10_for_ms) {
2016 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2019 memcpy(real_buffer + 8, buffer, len);
2023 real_buffer[2] = data->medium_type;
2024 real_buffer[3] = data->device_specific;
2025 real_buffer[4] = data->longlba ? 0x01 : 0;
2027 real_buffer[6] = data->block_descriptor_length >> 8;
2028 real_buffer[7] = data->block_descriptor_length;
2030 cmd[0] = MODE_SELECT_10;
2034 if (len > 255 || data->block_descriptor_length > 255 ||
2038 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2041 memcpy(real_buffer + 4, buffer, len);
2044 real_buffer[1] = data->medium_type;
2045 real_buffer[2] = data->device_specific;
2046 real_buffer[3] = data->block_descriptor_length;
2048 cmd[0] = MODE_SELECT;
2052 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2053 sshdr, timeout, retries, NULL);
2057 EXPORT_SYMBOL_GPL(scsi_mode_select);
2060 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2061 * @sdev: SCSI device to be queried
2062 * @dbd: set if mode sense will allow block descriptors to be returned
2063 * @modepage: mode page being requested
2064 * @buffer: request buffer (may not be smaller than eight bytes)
2065 * @len: length of request buffer.
2066 * @timeout: command timeout
2067 * @retries: number of retries before failing
2068 * @data: returns a structure abstracting the mode header data
2069 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2070 * must be SCSI_SENSE_BUFFERSIZE big.
2072 * Returns zero if unsuccessful, or the header offset (either 4
2073 * or 8 depending on whether a six or ten byte command was
2074 * issued) if successful.
2077 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2078 unsigned char *buffer, int len, int timeout, int retries,
2079 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2081 unsigned char cmd[12];
2084 int result, retry_count = retries;
2085 struct scsi_sense_hdr my_sshdr;
2087 memset(data, 0, sizeof(*data));
2088 memset(&cmd[0], 0, 12);
2090 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2091 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2094 /* caller might not be interested in sense, but we need it */
2099 use_10_for_ms = sdev->use_10_for_ms;
2101 if (use_10_for_ms) {
2105 cmd[0] = MODE_SENSE_10;
2112 cmd[0] = MODE_SENSE;
2117 memset(buffer, 0, len);
2119 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2120 sshdr, timeout, retries, NULL);
2122 /* This code looks awful: what it's doing is making sure an
2123 * ILLEGAL REQUEST sense return identifies the actual command
2124 * byte as the problem. MODE_SENSE commands can return
2125 * ILLEGAL REQUEST if the code page isn't supported */
2127 if (use_10_for_ms && !scsi_status_is_good(result) &&
2128 driver_byte(result) == DRIVER_SENSE) {
2129 if (scsi_sense_valid(sshdr)) {
2130 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2131 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2133 * Invalid command operation code
2135 sdev->use_10_for_ms = 0;
2141 if (scsi_status_is_good(result)) {
2142 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2143 (modepage == 6 || modepage == 8))) {
2144 /* Initio breakage? */
2147 data->medium_type = 0;
2148 data->device_specific = 0;
2150 data->block_descriptor_length = 0;
2151 } else if (use_10_for_ms) {
2152 data->length = buffer[0]*256 + buffer[1] + 2;
2153 data->medium_type = buffer[2];
2154 data->device_specific = buffer[3];
2155 data->longlba = buffer[4] & 0x01;
2156 data->block_descriptor_length = buffer[6]*256
2159 data->length = buffer[0] + 1;
2160 data->medium_type = buffer[1];
2161 data->device_specific = buffer[2];
2162 data->block_descriptor_length = buffer[3];
2164 data->header_length = header_length;
2165 } else if ((status_byte(result) == CHECK_CONDITION) &&
2166 scsi_sense_valid(sshdr) &&
2167 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2174 EXPORT_SYMBOL(scsi_mode_sense);
2177 * scsi_test_unit_ready - test if unit is ready
2178 * @sdev: scsi device to change the state of.
2179 * @timeout: command timeout
2180 * @retries: number of retries before failing
2181 * @sshdr: outpout pointer for decoded sense information.
2183 * Returns zero if unsuccessful or an error if TUR failed. For
2184 * removable media, UNIT_ATTENTION sets ->changed flag.
2187 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2188 struct scsi_sense_hdr *sshdr)
2191 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2195 /* try to eat the UNIT_ATTENTION if there are enough retries */
2197 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2199 if (sdev->removable && scsi_sense_valid(sshdr) &&
2200 sshdr->sense_key == UNIT_ATTENTION)
2202 } while (scsi_sense_valid(sshdr) &&
2203 sshdr->sense_key == UNIT_ATTENTION && --retries);
2207 EXPORT_SYMBOL(scsi_test_unit_ready);
2210 * scsi_device_set_state - Take the given device through the device state model.
2211 * @sdev: scsi device to change the state of.
2212 * @state: state to change to.
2214 * Returns zero if successful or an error if the requested
2215 * transition is illegal.
2218 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2220 enum scsi_device_state oldstate = sdev->sdev_state;
2222 if (state == oldstate)
2228 case SDEV_CREATED_BLOCK:
2239 case SDEV_TRANSPORT_OFFLINE:
2252 case SDEV_TRANSPORT_OFFLINE:
2260 case SDEV_TRANSPORT_OFFLINE:
2275 case SDEV_CREATED_BLOCK:
2284 case SDEV_CREATED_BLOCK:
2299 case SDEV_TRANSPORT_OFFLINE:
2311 case SDEV_TRANSPORT_OFFLINE:
2314 case SDEV_CREATED_BLOCK:
2322 sdev->offline_already = false;
2323 sdev->sdev_state = state;
2327 SCSI_LOG_ERROR_RECOVERY(1,
2328 sdev_printk(KERN_ERR, sdev,
2329 "Illegal state transition %s->%s",
2330 scsi_device_state_name(oldstate),
2331 scsi_device_state_name(state))
2335 EXPORT_SYMBOL(scsi_device_set_state);
2338 * sdev_evt_emit - emit a single SCSI device uevent
2339 * @sdev: associated SCSI device
2340 * @evt: event to emit
2342 * Send a single uevent (scsi_event) to the associated scsi_device.
2344 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2349 switch (evt->evt_type) {
2350 case SDEV_EVT_MEDIA_CHANGE:
2351 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2353 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2354 scsi_rescan_device(&sdev->sdev_gendev);
2355 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2357 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2358 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2360 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2361 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2363 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2364 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2366 case SDEV_EVT_LUN_CHANGE_REPORTED:
2367 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2369 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2370 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2372 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2373 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2382 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2386 * sdev_evt_thread - send a uevent for each scsi event
2387 * @work: work struct for scsi_device
2389 * Dispatch queued events to their associated scsi_device kobjects
2392 void scsi_evt_thread(struct work_struct *work)
2394 struct scsi_device *sdev;
2395 enum scsi_device_event evt_type;
2396 LIST_HEAD(event_list);
2398 sdev = container_of(work, struct scsi_device, event_work);
2400 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2401 if (test_and_clear_bit(evt_type, sdev->pending_events))
2402 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2405 struct scsi_event *evt;
2406 struct list_head *this, *tmp;
2407 unsigned long flags;
2409 spin_lock_irqsave(&sdev->list_lock, flags);
2410 list_splice_init(&sdev->event_list, &event_list);
2411 spin_unlock_irqrestore(&sdev->list_lock, flags);
2413 if (list_empty(&event_list))
2416 list_for_each_safe(this, tmp, &event_list) {
2417 evt = list_entry(this, struct scsi_event, node);
2418 list_del(&evt->node);
2419 scsi_evt_emit(sdev, evt);
2426 * sdev_evt_send - send asserted event to uevent thread
2427 * @sdev: scsi_device event occurred on
2428 * @evt: event to send
2430 * Assert scsi device event asynchronously.
2432 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2434 unsigned long flags;
2437 /* FIXME: currently this check eliminates all media change events
2438 * for polled devices. Need to update to discriminate between AN
2439 * and polled events */
2440 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2446 spin_lock_irqsave(&sdev->list_lock, flags);
2447 list_add_tail(&evt->node, &sdev->event_list);
2448 schedule_work(&sdev->event_work);
2449 spin_unlock_irqrestore(&sdev->list_lock, flags);
2451 EXPORT_SYMBOL_GPL(sdev_evt_send);
2454 * sdev_evt_alloc - allocate a new scsi event
2455 * @evt_type: type of event to allocate
2456 * @gfpflags: GFP flags for allocation
2458 * Allocates and returns a new scsi_event.
2460 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2463 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2467 evt->evt_type = evt_type;
2468 INIT_LIST_HEAD(&evt->node);
2470 /* evt_type-specific initialization, if any */
2472 case SDEV_EVT_MEDIA_CHANGE:
2473 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2474 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2475 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2476 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2477 case SDEV_EVT_LUN_CHANGE_REPORTED:
2478 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2479 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2487 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2490 * sdev_evt_send_simple - send asserted event to uevent thread
2491 * @sdev: scsi_device event occurred on
2492 * @evt_type: type of event to send
2493 * @gfpflags: GFP flags for allocation
2495 * Assert scsi device event asynchronously, given an event type.
2497 void sdev_evt_send_simple(struct scsi_device *sdev,
2498 enum scsi_device_event evt_type, gfp_t gfpflags)
2500 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2502 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2507 sdev_evt_send(sdev, evt);
2509 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2512 * scsi_device_quiesce - Block user issued commands.
2513 * @sdev: scsi device to quiesce.
2515 * This works by trying to transition to the SDEV_QUIESCE state
2516 * (which must be a legal transition). When the device is in this
2517 * state, only special requests will be accepted, all others will
2518 * be deferred. Since special requests may also be requeued requests,
2519 * a successful return doesn't guarantee the device will be
2520 * totally quiescent.
2522 * Must be called with user context, may sleep.
2524 * Returns zero if unsuccessful or an error if not.
2527 scsi_device_quiesce(struct scsi_device *sdev)
2529 struct request_queue *q = sdev->request_queue;
2533 * It is allowed to call scsi_device_quiesce() multiple times from
2534 * the same context but concurrent scsi_device_quiesce() calls are
2537 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2539 if (sdev->quiesced_by == current)
2544 blk_mq_freeze_queue(q);
2546 * Ensure that the effect of blk_set_pm_only() will be visible
2547 * for percpu_ref_tryget() callers that occur after the queue
2548 * unfreeze even if the queue was already frozen before this function
2549 * was called. See also https://lwn.net/Articles/573497/.
2552 blk_mq_unfreeze_queue(q);
2554 mutex_lock(&sdev->state_mutex);
2555 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2557 sdev->quiesced_by = current;
2559 blk_clear_pm_only(q);
2560 mutex_unlock(&sdev->state_mutex);
2564 EXPORT_SYMBOL(scsi_device_quiesce);
2567 * scsi_device_resume - Restart user issued commands to a quiesced device.
2568 * @sdev: scsi device to resume.
2570 * Moves the device from quiesced back to running and restarts the
2573 * Must be called with user context, may sleep.
2575 void scsi_device_resume(struct scsi_device *sdev)
2577 /* check if the device state was mutated prior to resume, and if
2578 * so assume the state is being managed elsewhere (for example
2579 * device deleted during suspend)
2581 mutex_lock(&sdev->state_mutex);
2582 if (sdev->quiesced_by) {
2583 sdev->quiesced_by = NULL;
2584 blk_clear_pm_only(sdev->request_queue);
2586 if (sdev->sdev_state == SDEV_QUIESCE)
2587 scsi_device_set_state(sdev, SDEV_RUNNING);
2588 mutex_unlock(&sdev->state_mutex);
2590 EXPORT_SYMBOL(scsi_device_resume);
2593 device_quiesce_fn(struct scsi_device *sdev, void *data)
2595 scsi_device_quiesce(sdev);
2599 scsi_target_quiesce(struct scsi_target *starget)
2601 starget_for_each_device(starget, NULL, device_quiesce_fn);
2603 EXPORT_SYMBOL(scsi_target_quiesce);
2606 device_resume_fn(struct scsi_device *sdev, void *data)
2608 scsi_device_resume(sdev);
2612 scsi_target_resume(struct scsi_target *starget)
2614 starget_for_each_device(starget, NULL, device_resume_fn);
2616 EXPORT_SYMBOL(scsi_target_resume);
2619 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2620 * @sdev: device to block
2622 * Pause SCSI command processing on the specified device. Does not sleep.
2624 * Returns zero if successful or a negative error code upon failure.
2627 * This routine transitions the device to the SDEV_BLOCK state (which must be
2628 * a legal transition). When the device is in this state, command processing
2629 * is paused until the device leaves the SDEV_BLOCK state. See also
2630 * scsi_internal_device_unblock_nowait().
2632 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2634 struct request_queue *q = sdev->request_queue;
2637 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2639 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2646 * The device has transitioned to SDEV_BLOCK. Stop the
2647 * block layer from calling the midlayer with this device's
2650 blk_mq_quiesce_queue_nowait(q);
2653 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2656 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2657 * @sdev: device to block
2659 * Pause SCSI command processing on the specified device and wait until all
2660 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2662 * Returns zero if successful or a negative error code upon failure.
2665 * This routine transitions the device to the SDEV_BLOCK state (which must be
2666 * a legal transition). When the device is in this state, command processing
2667 * is paused until the device leaves the SDEV_BLOCK state. See also
2668 * scsi_internal_device_unblock().
2670 static int scsi_internal_device_block(struct scsi_device *sdev)
2672 struct request_queue *q = sdev->request_queue;
2675 mutex_lock(&sdev->state_mutex);
2676 err = scsi_internal_device_block_nowait(sdev);
2678 blk_mq_quiesce_queue(q);
2679 mutex_unlock(&sdev->state_mutex);
2684 void scsi_start_queue(struct scsi_device *sdev)
2686 struct request_queue *q = sdev->request_queue;
2688 blk_mq_unquiesce_queue(q);
2692 * scsi_internal_device_unblock_nowait - resume a device after a block request
2693 * @sdev: device to resume
2694 * @new_state: state to set the device to after unblocking
2696 * Restart the device queue for a previously suspended SCSI device. Does not
2699 * Returns zero if successful or a negative error code upon failure.
2702 * This routine transitions the device to the SDEV_RUNNING state or to one of
2703 * the offline states (which must be a legal transition) allowing the midlayer
2704 * to goose the queue for this device.
2706 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2707 enum scsi_device_state new_state)
2709 switch (new_state) {
2711 case SDEV_TRANSPORT_OFFLINE:
2718 * Try to transition the scsi device to SDEV_RUNNING or one of the
2719 * offlined states and goose the device queue if successful.
2721 switch (sdev->sdev_state) {
2723 case SDEV_TRANSPORT_OFFLINE:
2724 sdev->sdev_state = new_state;
2726 case SDEV_CREATED_BLOCK:
2727 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2728 new_state == SDEV_OFFLINE)
2729 sdev->sdev_state = new_state;
2731 sdev->sdev_state = SDEV_CREATED;
2739 scsi_start_queue(sdev);
2743 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2746 * scsi_internal_device_unblock - resume a device after a block request
2747 * @sdev: device to resume
2748 * @new_state: state to set the device to after unblocking
2750 * Restart the device queue for a previously suspended SCSI device. May sleep.
2752 * Returns zero if successful or a negative error code upon failure.
2755 * This routine transitions the device to the SDEV_RUNNING state or to one of
2756 * the offline states (which must be a legal transition) allowing the midlayer
2757 * to goose the queue for this device.
2759 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2760 enum scsi_device_state new_state)
2764 mutex_lock(&sdev->state_mutex);
2765 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2766 mutex_unlock(&sdev->state_mutex);
2772 device_block(struct scsi_device *sdev, void *data)
2776 ret = scsi_internal_device_block(sdev);
2778 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2779 dev_name(&sdev->sdev_gendev), ret);
2783 target_block(struct device *dev, void *data)
2785 if (scsi_is_target_device(dev))
2786 starget_for_each_device(to_scsi_target(dev), NULL,
2792 scsi_target_block(struct device *dev)
2794 if (scsi_is_target_device(dev))
2795 starget_for_each_device(to_scsi_target(dev), NULL,
2798 device_for_each_child(dev, NULL, target_block);
2800 EXPORT_SYMBOL_GPL(scsi_target_block);
2803 device_unblock(struct scsi_device *sdev, void *data)
2805 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2809 target_unblock(struct device *dev, void *data)
2811 if (scsi_is_target_device(dev))
2812 starget_for_each_device(to_scsi_target(dev), data,
2818 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2820 if (scsi_is_target_device(dev))
2821 starget_for_each_device(to_scsi_target(dev), &new_state,
2824 device_for_each_child(dev, &new_state, target_unblock);
2826 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2829 scsi_host_block(struct Scsi_Host *shost)
2831 struct scsi_device *sdev;
2835 * Call scsi_internal_device_block_nowait so we can avoid
2836 * calling synchronize_rcu() for each LUN.
2838 shost_for_each_device(sdev, shost) {
2839 mutex_lock(&sdev->state_mutex);
2840 ret = scsi_internal_device_block_nowait(sdev);
2841 mutex_unlock(&sdev->state_mutex);
2843 scsi_device_put(sdev);
2849 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2850 * calling synchronize_rcu() once is enough.
2852 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2859 EXPORT_SYMBOL_GPL(scsi_host_block);
2862 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2864 struct scsi_device *sdev;
2867 shost_for_each_device(sdev, shost) {
2868 ret = scsi_internal_device_unblock(sdev, new_state);
2870 scsi_device_put(sdev);
2876 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2879 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2880 * @sgl: scatter-gather list
2881 * @sg_count: number of segments in sg
2882 * @offset: offset in bytes into sg, on return offset into the mapped area
2883 * @len: bytes to map, on return number of bytes mapped
2885 * Returns virtual address of the start of the mapped page
2887 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2888 size_t *offset, size_t *len)
2891 size_t sg_len = 0, len_complete = 0;
2892 struct scatterlist *sg;
2895 WARN_ON(!irqs_disabled());
2897 for_each_sg(sgl, sg, sg_count, i) {
2898 len_complete = sg_len; /* Complete sg-entries */
2899 sg_len += sg->length;
2900 if (sg_len > *offset)
2904 if (unlikely(i == sg_count)) {
2905 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2907 __func__, sg_len, *offset, sg_count);
2912 /* Offset starting from the beginning of first page in this sg-entry */
2913 *offset = *offset - len_complete + sg->offset;
2915 /* Assumption: contiguous pages can be accessed as "page + i" */
2916 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2917 *offset &= ~PAGE_MASK;
2919 /* Bytes in this sg-entry from *offset to the end of the page */
2920 sg_len = PAGE_SIZE - *offset;
2924 return kmap_atomic(page);
2926 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2929 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2930 * @virt: virtual address to be unmapped
2932 void scsi_kunmap_atomic_sg(void *virt)
2934 kunmap_atomic(virt);
2936 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2938 void sdev_disable_disk_events(struct scsi_device *sdev)
2940 atomic_inc(&sdev->disk_events_disable_depth);
2942 EXPORT_SYMBOL(sdev_disable_disk_events);
2944 void sdev_enable_disk_events(struct scsi_device *sdev)
2946 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2948 atomic_dec(&sdev->disk_events_disable_depth);
2950 EXPORT_SYMBOL(sdev_enable_disk_events);
2953 * scsi_vpd_lun_id - return a unique device identification
2954 * @sdev: SCSI device
2955 * @id: buffer for the identification
2956 * @id_len: length of the buffer
2958 * Copies a unique device identification into @id based
2959 * on the information in the VPD page 0x83 of the device.
2960 * The string will be formatted as a SCSI name string.
2962 * Returns the length of the identification or error on failure.
2963 * If the identifier is longer than the supplied buffer the actual
2964 * identifier length is returned and the buffer is not zero-padded.
2966 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2968 u8 cur_id_type = 0xff;
2970 const unsigned char *d, *cur_id_str;
2971 const struct scsi_vpd *vpd_pg83;
2972 int id_size = -EINVAL;
2975 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2982 * Look for the correct descriptor.
2983 * Order of preference for lun descriptor:
2984 * - SCSI name string
2985 * - NAA IEEE Registered Extended
2986 * - EUI-64 based 16-byte
2987 * - EUI-64 based 12-byte
2988 * - NAA IEEE Registered
2989 * - NAA IEEE Extended
2991 * as longer descriptors reduce the likelyhood
2992 * of identification clashes.
2995 /* The id string must be at least 20 bytes + terminating NULL byte */
3001 memset(id, 0, id_len);
3002 d = vpd_pg83->data + 4;
3003 while (d < vpd_pg83->data + vpd_pg83->len) {
3004 /* Skip designators not referring to the LUN */
3005 if ((d[1] & 0x30) != 0x00)
3008 switch (d[1] & 0xf) {
3011 if (cur_id_size > d[3])
3013 /* Prefer anything */
3014 if (cur_id_type > 0x01 && cur_id_type != 0xff)
3017 if (cur_id_size + 4 > id_len)
3018 cur_id_size = id_len - 4;
3020 cur_id_type = d[1] & 0xf;
3021 id_size = snprintf(id, id_len, "t10.%*pE",
3022 cur_id_size, cur_id_str);
3026 if (cur_id_size > d[3])
3028 /* Prefer NAA IEEE Registered Extended */
3029 if (cur_id_type == 0x3 &&
3030 cur_id_size == d[3])
3034 cur_id_type = d[1] & 0xf;
3035 switch (cur_id_size) {
3037 id_size = snprintf(id, id_len,
3042 id_size = snprintf(id, id_len,
3047 id_size = snprintf(id, id_len,
3058 if (cur_id_size > d[3])
3062 cur_id_type = d[1] & 0xf;
3063 switch (cur_id_size) {
3065 id_size = snprintf(id, id_len,
3070 id_size = snprintf(id, id_len,
3080 /* SCSI name string */
3081 if (cur_id_size + 4 > d[3])
3083 /* Prefer others for truncated descriptor */
3084 if (cur_id_size && d[3] > id_len)
3086 cur_id_size = id_size = d[3];
3088 cur_id_type = d[1] & 0xf;
3089 if (cur_id_size >= id_len)
3090 cur_id_size = id_len - 1;
3091 memcpy(id, cur_id_str, cur_id_size);
3092 /* Decrease priority for truncated descriptor */
3093 if (cur_id_size != id_size)
3106 EXPORT_SYMBOL(scsi_vpd_lun_id);
3109 * scsi_vpd_tpg_id - return a target port group identifier
3110 * @sdev: SCSI device
3112 * Returns the Target Port Group identifier from the information
3113 * froom VPD page 0x83 of the device.
3115 * Returns the identifier or error on failure.
3117 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3119 const unsigned char *d;
3120 const struct scsi_vpd *vpd_pg83;
3121 int group_id = -EAGAIN, rel_port = -1;
3124 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3130 d = vpd_pg83->data + 4;
3131 while (d < vpd_pg83->data + vpd_pg83->len) {
3132 switch (d[1] & 0xf) {
3134 /* Relative target port */
3135 rel_port = get_unaligned_be16(&d[6]);
3138 /* Target port group */
3139 group_id = get_unaligned_be16(&d[6]);
3148 if (group_id >= 0 && rel_id && rel_port != -1)
3153 EXPORT_SYMBOL(scsi_vpd_tpg_id);